Ben Hamner, CTO, Kaggle, NYC 2017
summary
Kaggle reports nearly one million members, says it has realized less than 1% of its potential, and announces acquisition by Google to accelerate growth. It built Kaggle Kernels reproducible notebooks and an open data platform Kaggle Datasets, which increase shared code examples roughly 100× and bundle data, code and metadata. Core claims: data access, messy labels, leakage and poor reproducibility are the biggest pain points; goal is one-command end-to-end reproduction, CI for data changes, and new competition formats like backtesting, live feeds and reinforcement learning.
AI-written summary · about
chapters (25)
- 0:00Intro
- 2:36Requirements
- 3:07Machine Learning
- 3:51Performance gains
- 4:14Avoiding overfitting
- 5:02Machine learning and teams
- 5:52Accessing data
- 6:08Highquality data
- 6:21Cleaning data
- 7:12Data leakage
- 8:11Going from research to production
- 8:35Reproducing work
- 9:37We can do better
- 9:56Repeat work
- 10:08Rewrite everything
- 10:43Broken stuff
- 11:20Helpful data
- 11:41Best practices
- 12:12Demo
- 15:59Collaboration
- 16:36Data Science
- 17:51Google
- 19:36Code competitions
- 19:48Competition formats
- 20:16Thank you
related talks
21mBen Hamner, Co-founder and CTO, Kaggle
12mCommunity matters: 8 reasons why you should be involved with
03mDBTB INT Anthony Goldbloom r
41mWhat Kaggle has learned from 2MM machine learning models
06mTop 5 takeaways
30mGraph Data Science
19mData science for real problems
26mReproducible and maintainable data science code with Kedro